Latent Class Moderated Mediation in Structural Equation Models: Applications and Limitations
Latent class moderated mediation in structural equation models can describe individual differences in psychological processes across latent groups. This method could be most useful where two latent classes have indirect paths of equal magnitude, but opposite signs. Global model fit indices often failed to detect misfit when a one-class model was fit to data from a two-class population with latent class moderated mediation. Under this misfitted one-class model, significant indirect effects were rarely found, even with strong indirect effects in each subpopulation. Information criteria only reliably selected the correct number of classes with a strong population indirect effect and measurement invariance. Violations of assumed strong measurement invariance led to selecting more classes than in the population.